An on-line scheduled multiple model/controller approach to nonlinear identification and control of a turbogenerator is presented. A local model network is used to represent the nonlinear dynamics of a turbogenerator. This comprises a nonlinear combination of local, linear submodels identified at different operating points. Nonlinear control of the turbogenerator by the automatic voltage regulator is then achieved by online blending of multiple PID controllers, each designed for a linear submodel. This approach has the practical advantage of being directly based on well-established principles from linear systems. The technique is tested on a validated simulation of a 3-kVA laboratory micro-machine system. The blending mechanism is derived from an analysis of the nonlinear characteristics of the system. Stability of the closed-loop system was proved using a stability theorem for Tanaka-Sugeno fuzzy systems along with a passivity stability principle. As expected, the resultant nonlinear multiple-controller automatic voltage regulator produces better overall performance than a single PID controller and a fixed-gain controller designed for a single operating point. The simulation studies further suggest that the new nonlinear controller can effectively handle variations in the operating point and is also tolerant to severe fault conditions.